Skip to content

// start

Platform Operator

You install and operate basebox Server on your own or dedicated hardware: bare metal, NVIDIA, Kubernetes, inference, monitoring. Your area is Developer → Installation & Operations.

This role exists only for basebox Server. In basebox Cloud, basebox handles operations completely; on the customer side only the Administrator exists there. On a server you are either the customer's IT or – if commissioned – basebox.

Typical tasks

Task Entry page
Understand the three layers and their resource needs Understand the architecture
Check hardware against a reference configuration Reference configurations
Prepare operating system, NVIDIA drivers, CUDA, Kubernetes Prepare the server · NVIDIA / GPU
Install basebox with Helm Using Helm charts · Install basebox
Place service models (RAG, OCR, STT) on GPU or CPU Deploy service models
Connect inference – bundled or on a separate GPU host Connect inference · Deployment topologies
Choose and size a language model Models & inference · LLM recommendations
Deploy connector services in the cluster MCP connectors with Helm
Connect Keycloak to LDAP / Active Directory Keycloak LDAP integration
Accept the installation Validate installation
Monitor, back up, update, scale Operations
Diagnose problems Troubleshooting · FAQ

What you are responsible for

On a self-managed server, everything operational is yours after handover of the software: operating system, drivers, Kubernetes, the basebox installation, service models, inference, network, backups, updates. basebox provides software, updates, documentation and support. If basebox is commissioned to operate, basebox takes over the agreed items via PAM/VPN or another agreed access path. The table: Responsibilities.

Not your job – in any model: application administration. Users, groups, apps, model selection and connector approvals belong to the administrator, even if you operate the server.

Three principles from operations

  1. basebox needs little GPU – inference and service models need it. Plan GPUs for the language model and for RAG/OCR/STT, not for the platform. Service models should not share the inference GPU.
  2. No silent fallback. If the inference endpoint is offline, basebox shows the error and recovers as soon as it is ready again. Monitor the endpoint accordingly.
  3. Back up before every upgrade. Above all the AISRV database; database migrations cannot be undone without a backup. See Updates.

For security and compliance questions

What is stored where on a server, what is encrypted, how remote maintenance works and what works in air-gapped operation: Security & Compliance → Server security.